Proceedings

Find matching any: Reset

Add filter to result:

Validation of Modicovi - Monocot and Dicot Coverage Ratio Vision Based Method for Real Time Estimation Canopy Coverage Ratio between Cereal Crops and Dicotyledon Weeds

H. S. Midtiby, R. N. Jørgensen, N. Krüger, M. S. Laursen

University of Southern Denmark

In agriculture, weed control is becoming an increasing problem due to limitations on the amount and on the number of herbicides permitted. This work presents a robust real time and leaf occlusion method capable of estimating the mono- and dicotyledon leaf coverage ratio within a given imaging area. The developed method (patent P1174DK00) estimates the coverage based on the shape of the vegetation within the image. It does this based on the edge of the segmented vegetation. The detected edges is divided up into a set of equally spaced points where each point is defined by the position of the edge point and the orientation of the gradient where the point was extracted.

For each point the relation to the neighboring points (points within a small distance typically ~6cm) is calculated. From the distribution of these relations, a set of descriptors are calculated which describe the shape. A descriptor may for example be an estimation of the average curvature in the image. In order to create a model for the relationship between the dicotyledons and the monocotyledons an artificial dataset was created. By using this dataset a non-linear regression model was created.

Preliminary simulated data where the estimated dicotyledon coverage was compared to a set of hand segmented images resulted in an estimation with a standard deviation of 8.9 percentage point.

The evaluation was performed on images taken of maize together with natural occurring weeds on images where the dicotyledon coverage ranged from 6% - 100% coverage and where 78% of the images contained partial occlusion. During spring 2012 performance of the algorithm will be tested on a larger in field dataset and it is the results of those tests which will be shown.

Keyword: soft objects, weeding, occlusion

H. S. Midtiby R. N. Jørgensen N. Krüger M. S. Laursen Machine Vision / Multispectral & Hyperspectral Imaging Applications to Precision Agriculture Poster 2012

Download paper